Private text to speech, with no cloud in the loop

Separate the speech engine from the rest of the app. Local synthesis avoids uploading your text for narration; it does not mean an application makes no network requests.

• Speech processing. LoudReader generates narration and recognises scanned text on your device.
• Account. No LoudReader account is required. That does not establish that an app collects no data.
• Diagnostics. Speech and text recognition run locally. The app also sends crash/performance diagnostics and usage analytics. Version 1.12 has no visible in-app switch for these diagnostics and analytics; website analytics consent is separate.
• Downloads. Getting a book, saving a web article or making a purchase uses a connection.

The privacy policy describes these flows.

It isn't malice, it's architecture. A cloud TTS service synthesizes speech on its own servers, which means your text has to travel there. That's unavoidable physics of the design: to hear a contract read aloud by a cloud voice, you send the contract. What happens next depends on each provider's terms (retention, logging, processing location), which you have to read, trust, and re-check every time they change.

On-device TTS removes the question instead of answering it. There is no speech-service copy of your document, but an app may still use external services for other purposes. For a novel this may not matter much; for a medical report, an unpublished manuscript, or a client's legal brief, it's the whole decision.

Download the voice resources, disconnect, and try a passage that has not already been rendered. That checks whether the app can generate new speech offline.

Offline playback alone is not proof of local synthesis: downloaded cloud audio also plays without a connection. It also says nothing about analytics sent when connectivity returns. Check the architecture and privacy disclosures separately.

For a concrete, fact-checked example of the cloud column (pricing, word metering, sign-in requirements), see the LoudReader vs Speechify comparison.

LoudReader turns any EPUB, PDF, or article into an audiobook with natural offline voices, on iPhone, iPad, and Apple Silicon Macs. Import a document and press play: a neural voice reads while each word highlights in the text, and your place is saved automatically. The free tier includes unlimited listening on every book, cover to cover, with no word quota. Private reading shouldn't be the expensive option. Its built-in catalog lets you browse and download 70,000+ Project Gutenberg titles, subject to local copyright, and the same on-device narration works for PDFs on your iPhone and articles on your Mac.

An honest accounting, because the trade-off is real:

• Languages and voice variety. Cloud services generally offer dozens of languages and huge voice catalogs. LoudReader has 23 natural offline narrators across 10 languages, plus any voice you clone from your own on the device.
• Hardware. Running a neural engine locally takes modern silicon: LoudReader needs an Apple Silicon Mac (macOS 15+) or an iPhone/iPad on iOS 18+. Cloud TTS runs on anything with a browser.
• Extras. Some cloud suites offer AI summaries and browser extensions. LoudReader focuses on reading and also includes on-device voice cloning, with a three-clone trial allowance.

Choose according to the documents, voices and controls you need. Local synthesis avoids a speech-provider upload; assess diagnostics, storage and your organisation’s requirements separately.